Qwen3.8-27B-BF16-NVLC-Lossless

Qwen3.8-27B BF16 NVLC Lossless

This repository contains a losslessly compressed representation of Qwen/Qwen3.8-27B at revision 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0.

• Implementation: TaimurAyaz/Qwen3.8-27B-BF16-NVLC
• Upstream model: Qwen/Qwen3.8-27B

NVLC is exact entropy coding. It does not quantize, round, retrain, or alter a BF16 weight. The integrated streaming verifier reconstructed every logical source shard and compared it byte-for-byte with the immutable upstream checkpoint.

Compression result

The repository contains only the canonical distribution payloads and their verification manifests. Local training samples, learned-model intermediates, build logs, and local command paths are intentionally omitted. These omitted files are not required to decode the checkpoint.

Encoding

Each BF16 word is separated into its exact fields:

[ sign:1 | exponent:8 | fraction:7 ]

The joint sign and exponent symbol is coded with a per-tensor rANS19 model. After that symbol is decoded, it selects a soft-routed fraction distribution for the exact seven-bit fraction. A small residual stream preserves the safetensors header and any non-BF16 spans. Reconstruction combines the exact fields bitwise.

Each source shard is represented by:

shards/model-NNNNN-of-00018/
  exponent-sign.esrans19
  fractions.srans19
  residual.nvlcpfx1
  shard-manifest.json

manifest.json records upstream provenance, artifact sizes and hashes, compression accounting, and all 18 exactness gates. SOURCE_SHA256SUMS records the expected hashes of fully reconstructed upstream safetensors shards.

DGX Spark result

The linked implementation derives a kernel-oriented Fixed7/Compact11 runtime from this canonical model. On one NVIDIA DGX Spark, TP1, C1, with four sequential 512-input to 256-output requests per repeat:

The frozen correctness request matched all 256 output token IDs, all 256 selected-token logprob rows, and the decoded text. The active runtime covered 851 of 851 required text tensors and retained no dense checkpoint or persistent decoded-weight cache.

The runtime result is narrow: one DGX Spark, language-model only, no MTP, concurrency one. Median mean TTFT increased from 635.617 ms to 1,255.411 ms in the sealed workload, while median mean TPOT decreased from 224.768 ms to 163.426 ms.

Download

hf download Taimur/Qwen3.8-27B-BF16-NVLC-Lossless \
  --local-dir qwen3.8-27b-bf16-nvlc

cd qwen3.8-27b-bf16-nvlc
shasum -a 256 -c SHA256SUMS

The NVLC files are not directly loadable by standard Transformers, vLLM, or an unmodified SGLang installation. Use the linked implementation to verify, reconstruct, or generate the DGX Spark runtime streams.

Exact reconstruction

Build the host reconstructor from the implementation repository, then run:

python3 reconstruct_stock_checkpoint.py \
  --compressed-dir qwen3.8-27b-bf16-nvlc \
  --output-dir /fresh/qwen3.8-27b-bf16-oracle \
  --stream-reconstructor build/bf16_stream_reconstruct \
  --expected-shards 18

The output path must not already exist. Reconstruction is bounded and streaming, but the complete dense output still requires approximately the upstream checkpoint size on disk. The compressed candidate runtime does not depend on this reconstructed output.

Runtime representation

The canonical rANS files are the distribution format. The measured GPU runtime derives fixed-width, lane-oriented streams once at initialization:

No GPU-native runtime cache is included in this model repository because it is hardware and kernel-layout specific. The implementation produces it from this canonical model and verifies each operation against the canonical source identity.

License

The model metadata and this compressed representation are distributed under the upstream Apache License 2.0. See LICENSE and UPSTREAM_README.md.